def main(args):
# read the description of the genetic code
with open(args.code) as fin_gcode:
arr = list(csv.reader(fin_gcode, delimiter='\t'))
indices, aminos, codons = zip(*arr)
if [int(x) for x in indices] != range(len(indices)):
raise ValueError
aminos = [x.lower() for x in aminos]
nstop = aminos.count('stop')
if nstop not in (2, 3, 4):
raise Exception('expected 2 or 3 or 4 stop codons')
if any(x == 'stop' for x in aminos[:-nstop]):
raise Exception('expected stop codons at the end of the genetic code')
# trim the stop codons
aminos = aminos[:-nstop]
codons = codons[:-nstop]
# precompute some numpy ndarrays using the genetic code
ts = design.get_nt_transitions(codons)
tv = design.get_nt_transversions(codons)
syn = design.get_syn(codons, aminos)
nonsyn = design.get_nonsyn(codons, aminos)
compo = design.get_compo(codons)
nt_sinks = design.get_nt_sinks(codons)
asym_compo = np.transpose(nt_sinks, (1, 2, 0))
# read the (nstates, nstates) array of observed codon substitutions
subs_counts = np.loadtxt(args.count_matrix, dtype=float)
# trim the stop codons
subs_counts = subs_counts[:-nstop, :-nstop]
# compute some summaries of the observed codon substitutions
counts = np.sum(subs_counts, axis=0) + np.sum(subs_counts, axis=1)
log_counts = np.log(counts)
empirical_codon_distn = counts / float(np.sum(counts))
# make a crude guess of the expected number of changes
log_blen = np.log(guess_branch_length(subs_counts))
# use the chosen model to construct an initial guess for max likelihood
model_natural_guess = args.model.get_natural_guess()
model_nparams = len(model_natural_guess)
encoded_guess = np.empty(model_nparams + 1, dtype=float)
encoded_guess[0] = log_blen
encoded_guess[1:] = args.model.natural_to_encoded(model_natural_guess)
# construct the neg log likelihood non-free params
neg_ll_args = (
args.model,
subs_counts,
log_counts, empirical_codon_distn,
ts, tv, syn, nonsyn, compo, asym_compo,
)
# define the objective function and the gradient and hessian
f_encoded_theta = functools.partial(
get_two_taxon_neg_ll_encoded_theta, *neg_ll_args)
g_encoded_theta = functools.partial(eval_grad, f_encoded_theta)
h_encoded_theta = functools.partial(eval_hess, f_encoded_theta)
# do the search, using information about the gradient and hessian
results = scipy.optimize.minimize(
f_encoded_theta,
encoded_guess,
method=args.minimization_method,
jac=g_encoded_theta,
hess=h_encoded_theta,
)
# extract and decode the maximum likelihood estimates
encoded_xopt = results.x
mle_log_blen = encoded_xopt[0]
mle_blen = np.exp(mle_log_blen)
model_encoded_xopt = encoded_xopt[1:]
model_xopt = args.model.encoded_to_natural(model_encoded_xopt)
xopt = np.empty_like(encoded_xopt)
xopt[0] = mle_blen
xopt[1:] = model_xopt
# check that the stationary distribution is ok
mle_distn = args.model.get_distn(
log_counts, empirical_codon_distn,
ts, tv, syn, nonsyn, compo, asym_compo,
model_xopt,
)
mle_pre_Q = args.model.get_pre_Q(
log_counts, empirical_codon_distn,
ts, tv, syn, nonsyn, compo, asym_compo,
model_xopt,
)
stationary_distn_check_helper(mle_pre_Q, mle_distn, mle_blen)
# define functions for computing the hessian
f = functools.partial(get_two_taxon_neg_ll, *neg_ll_args)
g = functools.partial(eval_grad, f)
h = functools.partial(eval_hess, f)
# report a summary of the maximum likelihood search
print 'raw results from the minimization:'
print results
print
print 'max likelihood branch length (expected number of substitutions):'
print mle_blen
print
print 'max likelihood estimates of other model parameters:'
print model_xopt
print
# print the hessian matrix at the max likelihood parameter values
fisher_info = h(xopt)
cov = scipy.linalg.inv(fisher_info)
errors = np.sqrt(np.diag(cov))
print 'observed fisher information matrix:'
print fisher_info
print
print 'inverse of fisher information matrix:'
print cov
print
print 'standard error estimates (sqrt of diag of inv of fisher info)'
print errors
print
# write the neg log likelihood into a separate file
if args.neg_log_likelihood_out:
with open(args.neg_log_likelihood_out, 'w') as fout:
print >> fout, results.fun
# write the parameter estimates into a separate file
if args.parameter_estimates_out:
with open(args.parameter_estimates_out, 'w') as fout:
for value in xopt:
print >> fout, value
# write the parameter estimates into a separate file
if args.parameter_errors_out:
with open(args.parameter_errors_out, 'w') as fout:
for value in errors:
print >> fout, value
def main(args):
# read the description of the genetic code
with open(args.code) as fin_gcode:
arr = list(csv.reader(fin_gcode, delimiter='\t'))
indices, aminos, codons = zip(*arr)
if [int(x) for x in indices] != range(len(indices)):
raise ValueError
aminos = [x.lower() for x in aminos]
nstop = aminos.count('stop')
if nstop not in (2, 3, 4):
raise Exception('expected 2 or 3 or 4 stop codons')
if any(x == 'stop' for x in aminos[:-nstop]):
raise Exception('expected stop codons at the end of the genetic code')
# trim the stop codons
aminos = aminos[:-nstop]
codons = codons[:-nstop]
# precompute some numpy ndarrays using the genetic code
ts = design.get_nt_transitions(codons)
tv = design.get_nt_transversions(codons)
syn = design.get_syn(codons, aminos)
nonsyn = design.get_nonsyn(codons, aminos)
compo = design.get_compo(codons)
nt_sinks = design.get_nt_sinks(codons)
asym_compo = np.transpose(nt_sinks, (1, 2, 0))
# read the (nstates, nstates) array of observed codon substitutions
subs_counts = np.loadtxt(args.count_matrix, dtype=float)
# trim the stop codons
subs_counts = subs_counts[:-nstop, :-nstop]
# compute some summaries of the observed codon substitutions
counts = np.sum(subs_counts, axis=0) + np.sum(subs_counts, axis=1)
log_counts = np.log(counts)
empirical_codon_distn = counts / float(np.sum(counts))
# make crude guesses about parameter values
blen = markovutil.guess_branch_length(subs_counts)
theta = args.model.get_natural_guess()
# Get the initial guesses for the EM parameters.
prior_probs = np.array([0.5, 0.5], dtype=float)
prior_em_distns = np.vstack((
empirical_codon_distn,
empirical_codon_distn,
))
# iteratively compute parameter estimates
for em_iteration_index in range(10):
# given parameter guesses, compute the pre-rate matrices
pre_Qs = args.model.get_pre_Qs(
prior_probs, prior_em_distns,
ts, tv, syn, nonsyn, compo, asym_compo,
theta)
# compute the appropriately scaled transition matrices
eq_distns = args.model.get_distns(
prior_probs, prior_em_distns,
ts, tv, syn, nonsyn, compo, asym_compo,
theta)
Ps = markovutil.get_branch_mix(
prior_probs, pre_Qs, eq_distns, blen)
# given parameter guesses, compute posterior expectations
post_probs, post_em_distns = get_posterior_expectations(
subs_counts, Ps, prior_probs, eq_distns)
# given posterior expectations, optimize the parameter guesses
encoded_theta = np.empty(len(theta) + 1, dtype=float)
encoded_theta[0] = np.log(blen)
encoded_theta[1:] = args.model.natural_to_encoded(theta)
# construct the neg log likelihood non-free params
neg_ll_args = (
args.model,
post_probs, post_em_distns,
subs_counts,
ts, tv, syn, nonsyn, compo, asym_compo,
)
# define the objective function and the gradient and hessian
f_encoded_theta = functools.partial(
get_two_taxon_neg_ll_encoded_theta, *neg_ll_args)
g_encoded_theta = functools.partial(eval_grad, f_encoded_theta)
h_encoded_theta = functools.partial(eval_hess, f_encoded_theta)
# do the search, using information about the gradient and hessian
results = scipy.optimize.minimize(
f_encoded_theta,
encoded_theta,
method=args.minimization_method,
jac=g_encoded_theta,
hess=h_encoded_theta,
)
# extract and decode the maximum likelihood estimates
encoded_xopt = results.x
mle_log_blen = encoded_xopt[0]
mle_blen = np.exp(mle_log_blen)
model_encoded_xopt = encoded_xopt[1:]
model_xopt = args.model.encoded_to_natural(
model_encoded_xopt)
xopt = np.empty_like(encoded_xopt)
xopt[0] = mle_blen
xopt[1:] = model_xopt
# report a summary of the maximum likelihood search
print 'raw results from the minimization:'
print results
print
print 'max likelihood branch length (expected number of substitutions):'
print mle_blen
print
print 'max likelihood estimates of other model parameters:'
print model_xopt
print
print 'posterior mixture probabilities:'
print post_probs
print
# get ready for the next iteration if we continue
blen = mle_blen
#theta = model_xopt
prior_probs = post_probs
prior_em_distns = post_em_distns
def main(args):
# read the description of the genetic code
with open(args.code) as fin_gcode:
arr = list(csv.reader(fin_gcode, delimiter='\t'))
indices, aminos, codons = zip(*arr)
if [int(x) for x in indices] != range(len(indices)):
raise ValueError
# look for stop codons
aminos = [x.lower() for x in aminos]
nstop = aminos.count('stop')
if nstop not in (2, 3, 4):
raise Exception('expected 2 or 3 or 4 stop codons')
if any(x == 'stop' for x in aminos[:-nstop]):
raise Exception('expected stop codons at the end of the genetic code')
# trim the stop codons
aminos = aminos[:-nstop]
codons = codons[:-nstop]
ncodons = len(codons)
# precompute some numpy ndarrays using the genetic code
ts = design.get_nt_transitions(codons)
tv = design.get_nt_transversions(codons)
syn = design.get_syn(codons, aminos)
nonsyn = design.get_nonsyn(codons, aminos)
compo = design.get_compo(codons)
nt_sinks = design.get_nt_sinks(codons)
asym_compo = np.transpose(nt_sinks, (1, 2, 0))
# read the empirical counts and detect and trim stop codon counts
empirical_codon_counts = np.loadtxt(args.empirical_codon_counts)
if len(empirical_codon_counts) < ncodons:
raise Exception
if any(empirical_codon_counts[ncodons:]):
raise Exception
empirical_codon_counts = empirical_codon_counts[:ncodons]
log_empirical_codon_counts = np.log(empirical_codon_counts)
empirical_codon_distn = empirical_codon_counts / float(
np.sum(empirical_codon_counts))
# Precompute the distribution over codon substitutions
# using the stationary distribution and transition matrix
# of the simulation model given the simulation parameters.
sim_theta = np.loadtxt(args.simulation_parameter_values)
log_sim_theta = np.log(sim_theta)
log_sim_blen = log_sim_theta[0]
log_sim_theta_model = log_sim_theta[1:]
sim_model = g_model_name_to_class[args.simulation_model]
pre_Q = sim_model.get_pre_Q(
log_empirical_codon_counts,
empirical_codon_distn,
ts, tv, syn, nonsyn, compo, asym_compo,
log_sim_theta_model,
)
stationary_distn = sim_model.get_distn(
log_empirical_codon_counts,
empirical_codon_distn,
ts, tv, syn, nonsyn, compo, asym_compo,
log_sim_theta_model,
)
# get the rate matrix
Q = markovutil.pre_Q_to_Q(pre_Q, stationary_distn, np.exp(log_sim_blen))
# get the conditional transition matrix
P = scipy.linalg.expm(Q)
# get the joint substitution probability matrix
J = (P.T * stationary_distn).T
# use an appropriate sample size
if args.sample_size:
sample_size = args.sample_size
else:
# If the sample size is unspecified,
# use a sample size whose number of codon counts
# matches the sum of the user-provided empirical codon counts.
sample_size = int(np.sum(empirical_codon_counts) / 2)
with open(args.table_out, 'w') as fout:
# write the header of the R table file
header_row = (
args.first_inference_model.replace('-', '.'),
args.second_inference_model.replace('-', '.'),
)
print >> fout, '\t'.join(header_row)
for sample_index in range(args.nsamples):
# sample an (ncodons, ncodons) ndarray of counts
# using the precomputed multinomial distribution
subs_counts = np.random.multinomial(
sample_size,
J.flat,
).reshape(J.shape).astype(float)
# compute the neg log likelihoods for the two models
neg_log_likelihoods = []
for model_name in (
args.first_inference_model,
args.second_inference_model,
):
model = g_model_name_to_class[model_name]
min_neg_ll = get_min_neg_log_likelihood(
model,
subs_counts,
ts, tv, syn, nonsyn, compo, asym_compo,
args.minimization_method,
)
neg_log_likelihoods.append(min_neg_ll)
# write the data row
data_row = (
sample_index+1,
-neg_log_likelihoods[0],
-neg_log_likelihoods[1],
)
print >> fout, '\t'.join(str(x) for x in data_row)
def main(args):
# read the description of the genetic code
with open(args.code_in) as fin_gcode:
arr = list(csv.reader(fin_gcode, delimiter='\t'))
indices, aminos, codons = zip(*arr)
if [int(x) for x in indices] != range(len(indices)):
raise ValueError
aminos = [x.lower() for x in aminos]
nstop = aminos.count('stop')
if nstop not in (2, 3, 4):
raise Exception('expected 2 or 3 or 4 stop codons')
if any(x == 'stop' for x in aminos[:-nstop]):
raise Exception('expected stop codons at the end of the genetic code')
# trim the stop codons
aminos = aminos[:-nstop]
codons = codons[:-nstop]
# load the ordered directed edges
DE = np.loadtxt(args.edges_in, delimiter='\t', dtype=int)
# load the alignment pattern
patterns = np.loadtxt(args.patterns_in, delimiter='\t', dtype=int)
# load the alignment weights
weights = np.loadtxt(args.weights_in, delimiter='\t', dtype=float)
# get the empirical codon distribution
ncodons = len(codons)
nsites = patterns.shape[0]
ntaxa = patterns.shape[1]
v_emp = np.zeros(ncodons, dtype=float)
for i in range(nsites):
for j in range(ntaxa):
state = patterns[i, j]
if state != -1:
v_emp[state] += weights[i]
v_emp /= np.sum(v_emp)
print 'empirical codon distribution:'
print v_emp
print
# precompute some design matrices
adj = design.get_adjacency(codons)
ts = design.get_nt_transitions(codons)
tv = design.get_nt_transversions(codons)
syn = design.get_syn(codons, aminos)
nonsyn = design.get_nonsyn(codons, aminos)
full_compo = design.get_full_compo(codons)
# For all of the data in the alignment,
# compute the grand total nucleotide counts at each of the three
# nucleotide positions within a codon.
# The full_compo ndarray has shape (ncodons, 3, 4)
# whereas the nucleotide distribution ndarray has shape (3, 4).
position_specific_nt_distns = np.tensordot(full_compo, v_emp, axes=(0,0))
# This is pre-computed if we want to use an empirical
# stationary distribution, but it is postponed if we want to use
# max likelihood parameters for the stationary distribution.
stationary_distn = codon1994.get_f3x4_codon_distn(
full_compo,
position_specific_nt_distns,
)
# construct the args to the neg log likelihood function
likelihood_args_empirical = (
patterns, weights,
stationary_distn,
ts, tv, syn, nonsyn,
)
likelihood_args_free = (
patterns, weights,
ts, tv, syn, nonsyn, full_compo,
)
# get the model A estimates using plain fmin
model_A_opt = scipy.optimize.fmin(
functools.partial(A.get_neg_ll, *likelihood_args_empirical),
A.get_guess(),
)
print 'optimal params for model A:'
print np.exp(model_A_opt)
print
# reconstruct the matrix
d = position_specific_nt_distns
lastcol = d[:, -1]
dratio = (d.T / lastcol).T
log_nt_guess = np.log(dratio[:, :-1]).reshape(9)
guess = np.hstack([model_A_opt, log_nt_guess])
# choose the model
likelihood_args = likelihood_args_free
#guess = get_guess_A_free
get_neg_ll = A_free.get_neg_ll
#likelihood_args = likelihood_args_empirical
#guess = theta_model_A
#get_neg_ll = get_neg_ll_model_A
# define the objective function and the gradient and hessian
f = functools.partial(get_neg_ll, *likelihood_args)
g = functools.partial(eval_grad, f)
h = functools.partial(eval_hess, f)
# do the search, using information about the gradient and hessian
"""
results = scipy.optimize.fmin_ncg(
f,
theta_model_A,
g,
fhess_p=None,
fhess=h,
avextol=1e-05,
epsilon=1.4901161193847656e-08,
maxiter=100,
full_output=True,
disp=1,
retall=0,
callback=None,
)
"""
#"""
results = scipy.optimize.fmin(
f,
guess,
)
#"""
# report the inital model results
print 'model A with empirical frequencies:'
print model_A_opt
print numpy.exp(model_A_opt)
print
# report a summary of the maximum likelihood search
print 'model A with free frequencies:'
print results
print numpy.exp(results)
print
def main(args):
# read the description of the genetic code
with open(args.code) as fin_gcode:
arr = list(csv.reader(fin_gcode, delimiter="\t"))
indices, aminos, codons = zip(*arr)
if [int(x) for x in indices] != range(len(indices)):
raise ValueError
aminos = [x.lower() for x in aminos]
nstop = aminos.count("stop")
if nstop not in (2, 3, 4):
raise Exception("expected 2 or 3 or 4 stop codons")
if any(x == "stop" for x in aminos[:-nstop]):
raise Exception("expected stop codons at the end of the genetic code")
# trim the stop codons
aminos = aminos[:-nstop]
codons = codons[:-nstop]
# precompute some numpy ndarrays using the genetic code
ts = design.get_nt_transitions(codons)
tv = design.get_nt_transversions(codons)
syn = design.get_syn(codons, aminos)
nonsyn = design.get_nonsyn(codons, aminos)
compo = design.get_compo(codons)
nt_sinks = design.get_nt_sinks(codons)
asym_compo = np.transpose(nt_sinks, (1, 2, 0))
# read the (nstates, nstates) array of observed codon substitutions
subs_counts = np.loadtxt(args.count_matrix, dtype=float)
# trim the stop codons
subs_counts = subs_counts[:-nstop, :-nstop]
# do the constrained log likelihood maximizations
min_lls = []
min_ll_slopes = []
junk_list = []
space = np.linspace(args.linspace_start, args.linspace_stop, num=args.linspace_num)
for kimura_d in space:
# define the model
model = FMutSelG_F_partial(kimura_d)
# compute the constrained min negative log likelihood
min_ll, min_ll_slope, xopt_full = get_min_neg_ll_and_slope_and_junk(
model, subs_counts, ts, tv, syn, nonsyn, compo, asym_compo, args.minimization_method
)
# add the min log likelihood to the list
min_lls.append(min_ll)
min_ll_slopes.append(min_ll_slope)
junk_list.append(xopt_full)
# write the R table
with open(args.table_out, "w") as fout:
# write the R header
header_row = ["Kimura.D", "min.neg.ll", "min.neg.ll.slope", "branch.length"] + FMutSelG_F.get_names()
print >> fout, "\t".join(header_row)
# write each row of the R table,
# where each row has
# position, kimura_d, min_ll
for i, v in enumerate(zip(space, min_lls, min_ll_slopes, junk_list)):
abc = list(v[:-1])
xopt = list(v[-1])
row = [i + 1] + abc + xopt
print >> fout, "\t".join(str(x) for x in row)
